Ho-Yin Mak

Infrastructure Planning for Electric Vehicles with Battery Swapping

The transportation sector is a major source of greenhouse gas (GHG) emissions. As a step toward a greener environment, solutions involving electric vehicles (EVs) have been proposed and discussed. When powered by electricity from efficient and environmentally-friendly generators, EVs have significantly lower per-mile running costs compared to gasoline cars, while generating lower emissions. Unfortunately, due to the limited capacity of batteries, typical EVs can only travel for about 100 miles on a single charge. Because recharging takes several hours, it is impossible to recharge an EV in the middle of a long (round) trip exceeding 100 miles. Better Place (BP), a start-up based in Palo Alto, CA, proposed a novel strategy that potentially overcomes the recharging problem. In the plan, in addition to charging adaptors at homes, work places and shopping malls, swapping stations"", at which depleted batteries can be exchanged for recharged ones in the middle of long trips, will be located at strategic locations along freeways. With its battery swapping equipment, BP has demonstrated how to effectively refuel an EV in less than two minutes. The possible success of EV solutions based on the idea of battery swapping hinges on the ability of the charging service provider (BP or other similar firms) to deploy a cost-effective infrastructure network with comprehensive coverage. Unfortunately, since the adoption rate of electric vehicles, and thus demand for swapping service, is still highly uncertain, the service provider must make deployment plans with incomplete information on hand. In this paper, we develop models that aid the planning process for deploying battery swapping infrastructure, based on a robust optimization framework. We further show that our models can be tightly approximated by mixed-integer second-order cone programs (MISOCPs), which are readily solvable by commercial solvers. Using these models, we demonstrate the potential impacts of battery standardization and various technology advancements on the optimal infrastructure deployment strategy.

Appointment Scheduling with Limited Distributional Information

In this paper, we develop distribution-free models that solve the appointment sequencing and scheduling problem by assuming only moments information of job durations. We show that our min--max appointment scheduling models, which minimize the worst-case expected waiting and overtime costs out of all probability distributions with the given marginal moments, can be exactly formulated as tractable conic programs. These formulations are obtained by exploiting hidden convexity of the problem. In the special case where only the first two marginal moments are given, the problem can be reformulated as a second-order cone program. Based on the structural properties of this formulation, under a mild condition, we derive the optimal time allowances in closed form and prove that it is optimal to sequence jobs in increasing order of job duration variance. We also prove similar results regarding the optimal time allowances and sequence for the case where only means and supports of job durations are known. This paper was accepted by Dimitris Bertsimas, optimization.

Toward Mass Adoption of Electric Vehicles: Impact of the Range and Resale Anxieties

Key to the mass adoption of electric vehicles (EVs) is the establishment of successful business models based on sound understanding of consumer behavior in adopting this new technology. In this paper, we study the impact of two major barriers to mass adoption of EVs: (i) range anxiety, the concern that the driving range of EVs may be insufficient to meet the driving needs, and (ii) resale anxiety, the concern that used values of EVs may deteriorate quickly. Using a stylized model calibrated to a data set based on the San Francisco Bay Area, we show that although both types of consumer anxieties typically harm the firm’s profit, they often improve consumer surplus. In addition, we show that a business model that requires consumers to lease the EV batteries (rather than purchase them) may lead to a greater level of adoption and emission savings when the level of resale anxiety is high. Further, a business model that offers EV range improvement through enhanced charging infrastructure typically yields greater adoption and consumer surplus, but lowers the firm’s profit, compared with one that offers enlarged batteries. Overall, we find that the combinations of battery owning/leasing with enhanced charging service, referred to as the (O, E) and (L, E) models in our paper, typically yield the best balance among the objectives of EV adoption, emission savings, profitability, and consumer surplus, when the degree of resale anxiety is low and high, respectively.

Risk diversification and risk pooling in supply chain design

Recent research has pointed out that the optimal strategies to mitigate supply disruptions and demand uncertainty are often mirror images of each other. In particular, risk diversification is favorable under the threat of disruptions and risk pooling is favorable under demand uncertainty. This article studies how dynamic sourcing in supply chain design provides partial benefits of both strategies. Optimization models are formulated for supply chain network design with dynamic sourcing under the risk of temporally dependent and temporally independent disruptions of facilities. Using computational experiments, it is shown that supply chain networks that allow small to moderate degrees of dynamic sourcing can be very robust against both disruptions and demand uncertainty. Insights are attained on the optimal degree of dynamic sourcing under different conditions.

Service Region Design for Urban Electric Vehicle Sharing Systems

Emerging collaborative consumption business models have shown promise in terms of both generating business opportunities and enhancing the efficient use of resources. In the transportation domain, car-sharing models are being adopted on a mass scale in major metropolitan areas worldwide. This mode of servicized mobility bridges the resource efficiency of public transit and the flexibility of personal transportation. Beyond the significant potential to reduce car ownership, car sharing shows promise in supporting the adoption of fuel-efficient vehicles, such as electric vehicles (EVs), because of these vehicles’ special cost structure with high purchase but low operating costs. Recently, key players in the car-sharing business, such as Autolib’, car2go, and DriveNow, have begun to employ EVs in an operations model that accommodates one-way trips. On the one hand (and particularly in free-floating car sharing), the one-way model results in significant improvements in coverage of travel needs and therefore i...

Sequencing Appointments for Service Systems Using Inventory Approximations

Managing appointments for service systems with random job durations is a challenging task. We consider a class of appointment planning problems that involve two sets of decisions: job sequencing , i.e., determining the order in which a list of jobs should be performed by the server, and appointment scheduling , i.e., planning the starting times for jobs. These decisions are interconnected because their joint goal is to minimize the expected server idle time and job late-start penalty costs incurred because of randomness in job durations. In this paper, we design new heuristics for sequencing appointments. The idea behind the development of these heuristics is the structural connection between such appointment scheduling problems and stochastic inventory control in serial supply chains. In particular, the decision of determining time allowances as buffers against random job durations is analogous to that of selecting inventory levels as buffers to accommodate random demand in a supply chain; having excess buffers in appointment scheduling and supply chain settings incurs idle time and excess inventory holding costs, respectively, and having inadequate buffers leads to delays of subsequent jobs and backorders, respectively. Recognizing this connection, we propose tractable approximations for the job sequencing problem, obtain several insights, and further develop a very simple sequencing rule of ordering jobs by duration variance to late-start penalty cost ratio. Computational results show that our proposed heuristics produce close-to-optimal job sequences with significantly reduced computation times compared with those produced using an exact mixed-integer stochastic programming formulation based on the sample-average approximation approach.

Alternative fuel station location model with demand learning

In this paper, we study the optimal location decision for a network of alternative fuel stations (AFS) servicing a new market where the demand rate for the refueling service can be learned over time. In the presence of demand learning, the firm needs to make a decision, whether to actively learn the market through a greater initial investment in the AFS network or defer the commitment since an overly-aggressive investment often results in sub-optimal AFS locations. To illustrate this trade-off, we introduce a two-stage location model, in which the service provider enters the market by deploying a set of stations in the first stage under uncertainty, and has the option to add more stations in the second stage after it learns the demand. The demand learning time (length of the first stage) is endogenously determined by the service provider’s action in the first stage. To solve this problem, we develop an efficient solution method that provides a framework to achieve a desired error rate of accuracy in the optimal solution. Using numerical experiment, we study the trade-off between active learning and deferred commitment in AFS deployment strategy under different market characteristics. Further, we find that the lack of planning foresight typically results in an over-commitment in facility investment while the service provider earns a lower expected profit.

Agility and Proximity Considerations in Supply Chain Design

Strategic supply chain design decisions are critical to the long-term success of a business. Traditional facility location models for supply chain design focus on the trade-offs between the costs and benefits of proximity, i.e., the distance between facilities and customers. These strategic-focused models do not consider the supply chains agility, i.e., its ability to quickly respond to unexpected fluctuations in customer needs. In this paper, we study the problem of designing a supply chain distribution network under demand uncertainty and analyze how the optimal design characteristics of proximity and agility depend on various input parameters. We are able to draw managerial insights on how agility considerations may invalidate well-established and widely-accepted qualitative results derived from traditional models. In particular, we show that it is optimal to increase the density of DCs when the shortage penalty cost increases, and to decrease the density of DCs when a certain unit transportation cost parameter increases. Through these findings, our work conveys the message that traditional, proximity-based facility location models can be inadequate for designing modern responsive supply chains, and calls for the need to develop a new class of models for the task.

Robust Optimization Approach to Empty Container Repositioning in Liner Shipping

In global container liner networks, the costly operations of empty container repositioning are necessitated by the imbalance of cargo flows across regions. Up to 40 and 60 % of containers shipped from Europe and North America to Asia are empty, respectively. Repositioning costs are sizable, often amounting up to 5–6 % of a shipping lines revenue. Therefore, identifying an optimal repositioning schedule to rebalance empty containers with minimal cost is one of the most critical planning problems in liner shipping. This is often complicated by the stochastic nature of demand and long transportation lead times. In this paper, we formulate a multiple-stage stochastic programming problem for the optimal repositioning of containers for a liner shipping network. As the problem is highly complex, the stochastic programming formulation is not computationally tractable. Therefore, we utilize emerging techniques in robust optimization to provide a tight approximation (bond) on the stochastic version of the problem. The resulting formulation is a second-order cone program (SOCP) and is computationally tractable. With this approximation, we perform computational experiments to evaluate the effectiveness of different repositioning policies.

Integrated Modeling for Location Analysis

Delivery of products and services relies on well-managed operations. In designing large-scaled supply chain and service systems, locations of key facilities are a critical decision, as these facilities form the backbone of operations of these systems. For example, a key to effective supply chain management is the deployment of a structurally well-designed facility network, consisting of plants, warehouses, retail stores, etc. The aim of the study of facility location is to develop analytical methodologies to inform the planning decisions for evaluating and selecting siting plans for these facilities that ensure both convenient provision of (or access to) products and services by customers and users, as well as efficient operations (i.e., low operating costs). design.